An empirical study of TensorFlow lite performances in Raspberry Pi / Mohamad Harith Mohd Zahid

Artificial intelligence today is essentially a machine that can do typically any specific task before mimicry to human intelligence. Its emphasis on a machine that thinks as it learns and experience without the help of human interaction. However, this type of learning is highly time-consuming and costly as very complex algorithms are present. Therefore, deep learning has been introduced and it has shown miraculous successes in the machine learning technique among a variety of functions. The popularity of its outcomes has open to several fields of studies with the help of deep learning open-source software tools in high powered devices. This research is proposed to predetermine the best suitable comparison of an empirical study of TensorFlow lite performances in Raspberry Pi. Therefore, with the use of deep learning open-source tools land pre-existed models from KERAS. The outcome of the benchmark is based on the throughput, energy, latency, memory footprint, and framerate per second in a low powered GPU device. Thus, it can be concluded that as the processing data increase the time average decrease but vice versa to the different types of models where it increases with the more data it processes over time.